Efficient Bayesian automatic calibration of a functional-structural wheat model using an adaptive design and a...

Area of Science:

• Plant modeling
• Computational biology
• Agricultural science

Background:

• Functional-structural plant models (FSPMs) are vital tools for plant science research.
• Model calibration is often hindered by high computational demands, leading to ignored error propagation.

Purpose of the Study:

• To develop and apply an automatic calibration method for the WALTer functional-structural wheat model.
• To address the computational challenges in calibrating complex plant models and quantify parameter uncertainty.

Main Methods:

• Utilized a Bayesian calibration approach to estimate five key parameters and their uncertainties.
• Employed Gaussian process metamodels to reduce the computational cost of the WALTer model.
• Implemented an adaptive design with an efficient global optimization algorithm for model calibration.

Main Results:

• The proposed method successfully calibrated the WALTer model using both synthetic and experimental data.
• Demonstrated the efficiency of Gaussian process metamodels in alleviating computational burden.
• Quantified parameter uncertainty effectively through the Bayesian framework.

Conclusions:

• The presented automatic calibration method is efficient for the WALTer wheat model.
• This approach offers a valuable solution for calibrating other complex functional-structural plant models.
• Reduces computational cost and improves accuracy in plant model parameterization.